--- gvpe/src/vpn_dns.C 2005/03/03 08:38:32 1.6 +++ gvpe/src/vpn_dns.C 2007/12/06 00:35:29 1.46 @@ -1,8 +1,10 @@ /* vpn_dns.C -- handle the dns tunnel part of the protocol. - Copyright (C) 2003-2004 Marc Lehmann + Copyright (C) 2003-2005 Marc Lehmann - This program is free software; you can redistribute it and/or modify + This file is part of GVPE. + + GVPE is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. @@ -13,10 +15,14 @@ GNU General Public License for more details. You should have received a copy of the GNU General Public License - along with this program; if not, write to the Free Software - Foundation, Inc. 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + along with gvpe; if not, write to the Free Software + Foundation, Inc. 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +// TODO: EDNS0 option to increase dns mtu? +// TODO: re-write dns packet parsing/creation using a safe mem-buffer +// to ensure no buffer overflows or similar problems. + #include "config.h" #if ENABLE_DNS @@ -25,6 +31,7 @@ // it's a hack, use only in emergency situations please. #include +#include #include #include @@ -37,126 +44,255 @@ #include +#include /* bug in libgmp: gmp.h relies on cstdio being included */ +#include + #include "netcompat.h" #include "vpn.h" -#define MIN_RETRY 1. -#define MAX_RETRY 60. +#define MAX_POLL_INTERVAL 5. // how often to poll minimally when the server has no data +#define ACTIVITY_INTERVAL 5. + +#define INITIAL_TIMEOUT 0.1 // retry timeouts +#define INITIAL_SYN_TIMEOUT 2. // retry timeout for initial syn + +#define MAX_SEND_INTERVAL 2. // optimistic? -#define MAX_OUTSTANDING 40 // max. outstanding requests -#define MAX_WINDOW 100 // max. for MAX_OUTSTANDING -#define MAX_RATE 1000 // requests/s -#define MAX_BACKLOG (10*1024) // size of protocol backlog, must be > MAXSIZE +#define MAX_WINDOW 1000 // max. for MAX_OUTSTANDING, and backlog +#define MAX_BACKLOG (64*1024) // size of gvpe protocol backlog (bytes), must be > MAXSIZE -#define MAX_DOMAIN_SIZE 220 // 255 is legal limit, but bind doesn't compress well +#define MAX_DOMAIN_SIZE 240 // 255 is legal limit, but bind doesn't compress well // 240 leaves about 4 bytes of server reply data -// every two request byte sless give room for one reply byte +// every request byte less give room for two reply bytes -// seqno has 12 bits, but the lower bit is always left as zero -// as bind caches ttl=0 records and we have to generate -// sequence numbers that always differ case-insensitively -#define SEQNO_MASK 0x07ff +#define SEQNO_MASK 0x3fff +#define SEQNO_EQ(a,b) ( 0 == ( ((a) ^ (b)) & SEQNO_MASK) ) -/* +#define MAX_LBL_SIZE 63 +#define MAX_PKT_SIZE 512 + +#define RR_TYPE_A 1 +#define RR_TYPE_NULL 10 +#define RR_TYPE_TXT 16 +#define RR_TYPE_ANY 255 -protocol, in shorthand :) +#define RR_CLASS_IN 1 -client -> server ANY? -server -> client TXT +#define CMD_IP_1 207 +#define CMD_IP_2 46 +#define CMD_IP_3 236 +#define CMD_IP_RST 29 +#define CMD_IP_SYN 113 +#define CMD_IP_REJ 32 - is dns64-encoded [] - is dns64-encoded <0:12>[] +// works for cmaps up to 255 (not 256!) +struct charmap +{ + enum { INVALID = (u8)255 }; -if is zero, the connection will be configured: + char encode [256]; // index => char + u8 decode [256]; // char => index + unsigned int size; -<0:12><0:4>client-id:12> + charmap (const char *cmap); +}; -*/ +charmap::charmap (const char *cmap) +{ + char *enc = encode; + u8 *dec = decode; -#define MAX_LBL_SIZE 63 -#define MAX_PKT_SIZE 512 + memset (enc, (char) 0, 256); + memset (dec, (char)INVALID, 256); -#define RR_TYPE_TXT 16 -#define RR_TYPE_ANY 255 -#define RR_CLASS_IN 1 + for (size = 0; cmap [size]; size++) + { + enc [size] = cmap [size]; + dec [(u8)enc [size]] = size; + } + + assert (size < 256); +} -// the "_" is not valid but widely accepted (all octets should be supported, but let's be conservative) -struct dns64 +#define MAX_DEC_LEN 500 +#define MAX_ENC_LEN (MAX_DEC_LEN * 2) +#define MAX_LIMBS ((MAX_DEC_LEN * 8 + GMP_NUMB_BITS - 1) / GMP_NUMB_BITS) + +// ugly. minimum base is 16(!) +struct basecoder { - static const char encode_chars[64 + 1]; - static s8 decode_chars[256]; + charmap cmap; + unsigned int enc_len [MAX_DEC_LEN]; + unsigned int dec_len [MAX_ENC_LEN]; - static int encode_len (int bytes) { return (bytes * 8 + 5) / 6; } - static int decode_len (int bytes) { return (bytes * 6) / 8; } - static int encode (char *dst, u8 *src, int len); - static int decode (u8 *dst, char *src, int len); + unsigned int encode_len (unsigned int len); + unsigned int decode_len (unsigned int len); - dns64 (); -} dns64; + unsigned int encode (char *dst, u8 *src, unsigned int len); + unsigned int decode (u8 *dst, char *src, unsigned int len); -// the following sequence has been crafted to -// a) look somewhat random -// b) the even (and odd) indices never share the same character as upper/lowercase -const char dns64::encode_chars[64 + 1] = "_-dDpPhHzZrR06QqMmjJkKBb34TtSsvVlL81xXaAeEFf92WwGgYyoO57UucCNniI"; -s8 dns64::decode_chars[256]; + basecoder (const char *cmap); +}; -dns64::dns64 () +basecoder::basecoder (const char *cmap) +: cmap (cmap) { - for (int i = 0; i < 64; i++) - decode_chars [encode_chars [i]] = i + 1; + for (unsigned int len = 0; len < MAX_DEC_LEN; ++len) + { + u8 src [MAX_DEC_LEN]; + u8 dst [MAX_ENC_LEN]; + + memset (src, 255, len); + + mp_limb_t m [MAX_LIMBS]; + mp_size_t n; + + n = mpn_set_str (m, src, len, 256); + n = mpn_get_str (dst, this->cmap.size, m, n); + + for (int i = 0; !dst [i]; ++i) + n--; + + enc_len [len] = n; + dec_len [n] = len; + } } -int dns64::encode (char *dst, u8 *src, int len) +unsigned int basecoder::encode_len (unsigned int len) { - // slow, but easy to debug - char *beg = dst; - unsigned int accum, bits = 0; + return enc_len [len]; +} - while (len--) - { - accum <<= 8; - accum |= *src++; - bits += 8; +unsigned int basecoder::decode_len (unsigned int len) +{ + while (len && !dec_len [len]) + --len; - while (bits >= 6) - { - *dst++ = encode_chars [(accum >> (bits - 6)) & 63]; - bits -= 6; - } + return dec_len [len]; +} + +unsigned int basecoder::encode (char *dst, u8 *src, unsigned int len) +{ + if (!len || len > MAX_DEC_LEN) + return 0; + + int elen = encode_len (len); + + mp_limb_t m [MAX_LIMBS]; + mp_size_t n; + + u8 dst_ [MAX_ENC_LEN]; + + n = mpn_set_str (m, src, len, 256); + n = mpn_get_str (dst_, cmap.size, m, n); + + int plen = elen; // for padding + + while (n < plen) + { + *dst++ = cmap.encode [0]; + plen--; } - if (bits) - *dst++ = encode_chars [(accum << (6 - bits)) & 63]; + for (unsigned int i = n - plen; i < n; ++i) + *dst++ = cmap.encode [dst_ [i]]; - return dst - beg; + return elen; } -int dns64::decode (u8 *dst, char *src, int len) +unsigned int basecoder::decode (u8 *dst, char *src, unsigned int len) { - // slow, but easy to debug - u8 *beg = dst; - unsigned int accum, bits = 0; + if (!len || len > MAX_ENC_LEN) + return 0; + + u8 src_ [MAX_ENC_LEN]; + unsigned int elen = 0; while (len--) { - s8 chr = decode_chars [(u8)*src++]; + u8 val = cmap.decode [(u8)*src++]; - if (!chr) - continue; + if (val != charmap::INVALID) + src_ [elen++] = val; + } - accum <<= 6; - accum |= chr - 1; - bits += 6; + int dlen = decode_len (elen); - while (bits >= 8) - { - *dst++ = accum >> (bits - 8); - bits -= 8; - } + mp_limb_t m [MAX_LIMBS]; + mp_size_t n; + + u8 dst_ [MAX_DEC_LEN]; + + n = mpn_set_str (m, src_, elen, cmap.size); + n = mpn_get_str (dst_, 256, m, n); + + if (n < dlen) + { + memset (dst, 0, dlen - n); + memcpy (dst + dlen - n, dst_, n); } + else + memcpy (dst, dst_ + n - dlen, dlen); + + return dlen; +} + +#if 0 +struct test { test (); } test; + +test::test () +{ + basecoder cdc ("0123456789abcdefghijklmnopqrstuvwxyz"); + + u8 in[] = "0123456789abcdefghijklmnopqrstuvwxyz"; + static char enc[200]; + static u8 dec[200]; + + for (int i = 1; i < 20; i++) + { + int elen = cdc.encode (enc, in, i); + int dlen = cdc.decode (dec, enc, elen); + + printf ("%d>%d>%d (%s>%s)\n", i, elen, dlen, enc, dec); + } + abort (); +} +#endif + +//static basecoder cdc64 ("_dDpPhHzZrR06QqMmjJkKBb34TtSsvVlL81xXaAeEFf92WwGgYyoO57UucCNniI-"); +//static basecoder cdc63 ("_dDpPhHzZrR06QqMmjJkKBb34TtSsvVlL81xXaAeEFf92WwGgYyoO57UucCNniI"); +static basecoder cdc62 ("dDpPhHzZrR06QqMmjJkKBb34TtSsvVlL81xXaAeEFf92WwGgYyoO57UucCNniI"); +//static basecoder cdc36 ("dphzr06qmjkb34tsvl81xaef92wgyo57ucni"); // unused as of yet +static basecoder cdc26 ("dPhZrQmJkBtSvLxAeFwGyO"); + +///////////////////////////////////////////////////////////////////////////// + +#define HDRSIZE 6 + +inline void encode_header (char *data, int clientid, int seqno, int retry = 0) +{ + seqno &= SEQNO_MASK; - return dst - beg; + u8 hdr[3] = { + clientid, + (seqno >> 8) | (retry << 6), + seqno, + }; + + assert (clientid < 256); + + cdc26.encode (data, hdr, 3); +} + +inline void decode_header (char *data, int &clientid, int &seqno) +{ + u8 hdr[3]; + + cdc26.decode (hdr, data, HDRSIZE); + + clientid = hdr[0]; + seqno = ((hdr[1] << 8) | hdr[2]) & SEQNO_MASK; } ///////////////////////////////////////////////////////////////////////////// @@ -195,7 +331,7 @@ void byte_stream::remove (int count) { if (count > fill) - abort (); + assert (count <= fill); memmove (data, data + count, fill -= count); } @@ -218,25 +354,34 @@ data [fill++] = pkt->len >> 8; data [fill++] = pkt->len; - memcpy (data + fill, &((*pkt)[0]), pkt->len); fill += pkt->len; + memcpy (data + fill, pkt->at (0), pkt->len); fill += pkt->len; return true; } vpn_packet *byte_stream::get () { - int len = (data [0] << 8) | data [1]; + unsigned int len; + + for (;;) + { + len = (data [0] << 8) | data [1]; - if (len > MAXSIZE && fill >= 2) - abort (); // TODO handle this gracefully, connection reset + if (len <= MAXSIZE || fill < 2) + break; + // TODO: handle this better than skipping, e.g. by reset + slog (L_DEBUG, _("DNS: corrupted packet stream skipping a byte...")); + remove (1); + } + if (fill < len + 2) return 0; vpn_packet *pkt = new vpn_packet; pkt->len = len; - memcpy (&((*pkt)[0]), data + 2, len); + memcpy (pkt->at (0), data + 2, len); remove (len + 2); return pkt; @@ -246,7 +391,7 @@ #define FLAG_QUERY ( 0 << 15) #define FLAG_RESPONSE ( 1 << 15) -#define FLAG_OP_MASK (15 << 14) +#define FLAG_OP_MASK (15 << 11) #define FLAG_OP_QUERY ( 0 << 11) #define FLAG_AA ( 1 << 10) #define FLAG_TC ( 1 << 9) @@ -263,13 +408,81 @@ #define DEFAULT_CLIENT_FLAGS (FLAG_QUERY | FLAG_OP_QUERY | FLAG_RD) #define DEFAULT_SERVER_FLAGS (FLAG_RESPONSE | FLAG_OP_QUERY | FLAG_AA | FLAG_RD | FLAG_RA) +struct dns_cfg +{ + static int next_uid; + + u8 id1, id2, id3, id4; + + u8 version; + u8 flags; + u8 rrtype; + u8 def_ttl; + + u16 client; + u16 uid; // to make request unique + + u16 max_size; + u8 seq_cdc; + u8 req_cdc; + + u8 rep_cdc; + u8 delay; // time in 0.01s units that the server may delay replying packets + u8 r3, r4; + + u8 r5, r6, r7, r8; + + void reset (int clientid); + bool valid (); +}; + +int dns_cfg::next_uid; + +void dns_cfg::reset (int clientid) +{ + id1 = 'G'; + id2 = 'V'; + id3 = 'P'; + id4 = 'E'; + + version = 1; + + rrtype = RR_TYPE_TXT; + flags = 0; + def_ttl = 0; + seq_cdc = 26; + req_cdc = 62; + rep_cdc = 0; + max_size = htons (MAX_PKT_SIZE); + client = htons (clientid); + uid = next_uid++; + delay = 0; + + r3 = r4 = 0; + r4 = r5 = r6 = r7 = 0; +} + +bool dns_cfg::valid () +{ + // although the protocol itself allows for some configurability, + // only the following encoding/decoding settings are implemented. + return id1 == 'G' + && id2 == 'V' + && id3 == 'P' + && id4 == 'E' + && seq_cdc == 26 + && req_cdc == 62 + && rep_cdc == 0 + && version == 1; +} + struct dns_packet : net_packet { u16 id; u16 flags; // QR:1 Opcode:4 AA:1 TC:1 RD:1 RA:1 Z:3 RCODE:4 u16 qdcount, ancount, nscount, arcount; - u8 data[MAXSIZE - 6 * 2]; + u8 data [MAXSIZE - 6 * 2]; int decode_label (char *data, int size, int &offs); }; @@ -310,22 +523,13 @@ ///////////////////////////////////////////////////////////////////////////// -struct dns_req -{ - dns_packet *pkt; - tstamp next; - int retry; - connection *conn; - int seqno; - - dns_req (connection *c); - void gen_stream_req (int seqno, byte_stream *stream); -}; - -static u16 dns_id = 12098; // TODO: should be per-vpn +static u16 dns_id = 0; // TODO: should be per-vpn static u16 next_id () { + if (!dns_id) + dns_id = time (0); + // the simplest lsfr with periodicity 65535 i could find dns_id = (dns_id << 1) | (((dns_id >> 1) @@ -336,45 +540,118 @@ return dns_id; } -dns_req::dns_req (connection *c) -: conn (c) +struct dns_rcv; +struct dns_snd; + +struct dns_connection +{ + connection *c; + struct vpn *vpn; + + dns_cfg cfg; + + bool established; + + tstamp last_received; + tstamp last_sent; + double min_latency; + double poll_interval, send_interval; + + vector rcvpq; + + byte_stream rcvdq; int rcvseq; int repseq; + byte_stream snddq; int sndseq; + + inline void time_cb (ev::timer &w, int revents); ev::timer tw; + void receive_rep (dns_rcv *r); + + dns_connection (connection *c); + ~dns_connection (); +}; + +struct dns_snd +{ + dns_packet *pkt; + tstamp timeout, sent; + int retry; + struct dns_connection *dns; + int seqno; + bool stdhdr; + + void gen_stream_req (int seqno, byte_stream &stream); + void gen_syn_req (); + + dns_snd (dns_connection *dns); + ~dns_snd (); +}; + +dns_snd::dns_snd (dns_connection *dns) +: dns (dns) { - next = 0; + timeout = 0; retry = 0; + seqno = 0; + sent = ev_now (); + stdhdr = false; pkt = new dns_packet; pkt->id = next_id (); } -void dns_req::gen_stream_req (int seqno, byte_stream *stream) +dns_snd::~dns_snd () { + delete pkt; +} + +static void append_domain (dns_packet &pkt, int &offs, const char *domain) +{ + // add tunnel domain + for (;;) + { + const char *end = strchr (domain, '.'); + + if (!end) + end = domain + strlen (domain); + + int len = end - domain; + + pkt [offs++] = len; + memcpy (pkt.at (offs), domain, len); + offs += len; + + if (!*end) + break; + + domain = end + 1; + } +} + +void dns_snd::gen_stream_req (int seqno, byte_stream &stream) +{ + stdhdr = true; this->seqno = seqno; + timeout = ev_now () + INITIAL_TIMEOUT; + pkt->flags = htons (DEFAULT_CLIENT_FLAGS); pkt->qdcount = htons (1); int offs = 6*2; - int dlen = MAX_DOMAIN_SIZE - strlen (THISNODE->domain) - 2; + int dlen = MAX_DOMAIN_SIZE - (strlen (dns->c->conf->domain) + 2); // MAX_DOMAIN_SIZE is technically 255, but bind doesn't compress responses well, // so we need to have space for 2*MAX_DOMAIN_SIZE + header + extra - u8 data[256]; //TODO - - data[0] = THISNODE->id; //TODO - data[1] = seqno >> 7; //TODO - data[2] = seqno << 1; //TODO + char enc[256], *encp = enc; + encode_header (enc, THISNODE->id, seqno); - int datalen = dns64::decode_len (dlen - (dlen + MAX_LBL_SIZE - 1) / MAX_LBL_SIZE) - 3; + int datalen = cdc62.decode_len (dlen - (dlen + MAX_LBL_SIZE - 1) / MAX_LBL_SIZE - HDRSIZE); - if (datalen > stream->size ()) - datalen = stream->size (); + if (datalen > stream.size ()) + datalen = stream.size (); - char enc[256], *encp = enc; - - memcpy (data + 3, stream->begin (), datalen); - int enclen = dns64::encode (enc, data, datalen + 3); - stream->remove (datalen); + int enclen = cdc62.encode (enc + HDRSIZE, stream.begin (), datalen) + HDRSIZE; + stream.remove (datalen); while (enclen) { @@ -389,30 +666,34 @@ enclen -= lbllen; } - const char *suffix = THISNODE->domain; + append_domain (*pkt, offs, dns->c->conf->domain); - // add tunnel domain - for (;;) - { - const char *end = strchr (suffix, '.'); + (*pkt)[offs++] = 0; + (*pkt)[offs++] = RR_TYPE_ANY >> 8; (*pkt)[offs++] = RR_TYPE_ANY; + (*pkt)[offs++] = RR_CLASS_IN >> 8; (*pkt)[offs++] = RR_CLASS_IN; - if (!end) - end = suffix + strlen (suffix); + pkt->len = offs; +} - int len = end - suffix; +void dns_snd::gen_syn_req () +{ + timeout = ev_now () + INITIAL_SYN_TIMEOUT; - (*pkt)[offs++] = len; - memcpy (&((*pkt)[offs]), suffix, len); - offs += len; + pkt->flags = htons (DEFAULT_CLIENT_FLAGS); + pkt->qdcount = htons (1); - if (!*end) - break; + int offs = 6 * 2; - suffix = end + 1; - } + int elen = cdc26.encode ((char *)pkt->at (offs + 1), (u8 *)&dns->cfg, sizeof (dns_cfg)); + + assert (elen <= MAX_LBL_SIZE); + + (*pkt)[offs] = elen; + offs += elen + 1; + append_domain (*pkt, offs, dns->c->conf->domain); (*pkt)[offs++] = 0; - (*pkt)[offs++] = RR_TYPE_ANY >> 8; (*pkt)[offs++] = RR_TYPE_ANY; + (*pkt)[offs++] = RR_TYPE_A >> 8; (*pkt)[offs++] = RR_TYPE_A; (*pkt)[offs++] = RR_CLASS_IN >> 8; (*pkt)[offs++] = RR_CLASS_IN; pkt->len = offs; @@ -441,370 +722,593 @@ } ///////////////////////////////////////////////////////////////////////////// + +dns_connection::dns_connection (connection *c) +: c (c) +, rcvdq (MAX_BACKLOG * 2) +, snddq (MAX_BACKLOG) +{ + tw.set (this); -struct dns_cfg + vpn = c->vpn; + + established = false; + + rcvseq = repseq = sndseq = 0; + + last_sent = last_received = 0; + poll_interval = 0.5; // starting here + send_interval = 0.5; // starting rate + min_latency = INITIAL_TIMEOUT; +} + +dns_connection::~dns_connection () { - u8 id1, id2, id3; - u8 def_ttl; - u8 unused1; - u16 max_size; - u8 flags1, flags2; -}; + for (vector::iterator i = rcvpq.begin (); + i != rcvpq.end (); + ++i) + delete *i; +} -void connection::dnsv4_receive_rep (struct dns_rcv *r) +void dns_connection::receive_rep (dns_rcv *r) { - dns_rcvpq.push_back (r); + if (r->datalen) + { + last_received = ev_now (); + tw (); + + poll_interval = send_interval; + } + else + { + poll_interval *= 1.5; + + if (poll_interval > MAX_POLL_INTERVAL) + poll_interval = MAX_POLL_INTERVAL; + } + + rcvpq.push_back (r); redo: - for (vector::iterator i = dns_rcvpq.begin (); - i != dns_rcvpq.end (); - ++i) - if (dns_rcvseq == (*i)->seqno) + // find next packet + for (vector::iterator i = rcvpq.end (); i-- != rcvpq.begin (); ) + if (SEQNO_EQ (rcvseq, (*i)->seqno)) { - dns_rcv *r = *i; + //printf ("seqno eq %x %x\n", rcvseq, (*i)->seqno);//D + // enter the packet into our input stream + r = *i; + + // remove the oldest packet, look forward, as it's oldest first + for (vector::iterator j = rcvpq.begin (); j != rcvpq.end (); ++j) + if (SEQNO_EQ ((*j)->seqno, rcvseq - MAX_WINDOW)) + { + //printf ("seqno RR %x %x\n", (*j)->seqno, rcvseq - MAX_WINDOW);//D + delete *j; + rcvpq.erase (j); + break; + } - dns_rcvseq = (dns_rcvseq + 1) & SEQNO_MASK; + rcvseq = (rcvseq + 1) & SEQNO_MASK; - if (!dns_snddq && !dns_rcvdq) + if (!rcvdq.put (r->data, r->datalen)) { - dns_rcvdq = new byte_stream (MAX_BACKLOG * 2); - dns_snddq = new byte_stream (MAX_BACKLOG); - - dns_si.set (::conf.dns_forw_host, ::conf.dns_forw_port, PROT_DNSv4); + slog (L_ERR, "DNS: !rcvdq.put (r->data, r->datalen)"); + abort (); // MUST never overflow, can be caused by data corruption, TODO } - if (!dns_rcvdq->put (r->data, r->datalen)) - abort (); // MUST never overflow, can be caused by data corruption, TODO - - while (vpn_packet *pkt = dns_rcvdq->get ()) + while (vpn_packet *pkt = rcvdq.get ()) { sockinfo si; - si.host = 0; si.port = 0; si.prot = PROT_DNSv4; + si.host = htonl (c->conf->id); si.port = 0; si.prot = PROT_DNSv4; vpn->recv_vpn_packet (pkt, si); + + delete pkt; } - } - else if ((u32)(*i)->seqno - (u32)dns_rcvseq + MAX_WINDOW > MAX_WINDOW * 2) - { - //D - //abort(); - printf ("%d erasing %d (%d)\n", THISNODE->id, (u32)(*i)->seqno, dns_rcvseq); - dns_rcvpq.erase (i); + + // check for further packets goto redo; } } -dns_packet * -vpn::dnsv4_server (dns_packet *pkt) +void +vpn::dnsv4_server (dns_packet &pkt) { - u16 flags = ntohs (pkt->flags); + u16 flags = ntohs (pkt.flags); - //memcpy (&((*rep)[0]), &((*pkt)[0]), pkt->len); int offs = 6 * 2; // skip header - pkt->flags = htons (DEFAULT_SERVER_FLAGS | FLAG_RCODE_FORMERR); + pkt.flags = htons (DEFAULT_SERVER_FLAGS | FLAG_RCODE_FORMERR); - if (!(flags & (FLAG_RESPONSE | FLAG_OP_MASK | FLAG_TC)) - && pkt->qdcount == htons (1)) + if (0 == (flags & (FLAG_RESPONSE | FLAG_OP_MASK)) + && pkt.qdcount == htons (1)) { - char qname[MAXSIZE]; - int qlen = pkt->decode_label ((char *)qname, MAXSIZE - offs, offs); + char qname [MAXSIZE]; + int qlen = pkt.decode_label ((char *)qname, MAXSIZE - offs, offs); - u16 qtype = (*pkt) [offs++] << 8; qtype |= (*pkt) [offs++]; - u16 qclass = (*pkt) [offs++] << 8; qclass |= (*pkt) [offs++]; + u16 qtype = pkt [offs++] << 8; qtype |= pkt [offs++]; + u16 qclass = pkt [offs++] << 8; qclass |= pkt [offs++]; - pkt->qdcount = htons (1); - pkt->ancount = 0; - pkt->nscount = 0; // should be self, as other nameservers reply like this - pkt->arcount = 0; // a record for self, as other nameservers reply like this + pkt.qdcount = htons (1); + pkt.ancount = 0; + pkt.nscount = 0; // should be self, as other nameservers reply like this + pkt.arcount = 0; // a record for self, as other nameservers reply like this - pkt->flags = htons (DEFAULT_SERVER_FLAGS | FLAG_RCODE_NXDOMAIN); + pkt.flags = htons (DEFAULT_SERVER_FLAGS | FLAG_RCODE_SERVFAIL); int dlen = strlen (THISNODE->domain); if (qclass == RR_CLASS_IN - && (qtype == RR_TYPE_ANY || qtype == RR_TYPE_TXT) && qlen > dlen + 1 - && !memcmp (qname + qlen - dlen - 1, THISNODE->domain, dlen)) + && !memcmp (qname + qlen - (dlen + 1), THISNODE->domain, dlen)) { - // correct class, domain: parse - u8 data[MAXSIZE]; - int datalen = dns64::decode (data, qname, qlen - dlen - 1); - - int client = data[0]; - int seqno = ((data[1] << 7) | (data[2] >> 1)) & SEQNO_MASK; - - if (0 < client && client <= conns.size ()) + // now generate reply + pkt.ancount = htons (1); // one answer RR + pkt.flags = htons (DEFAULT_SERVER_FLAGS | FLAG_RCODE_OK); + + if ((qtype == RR_TYPE_ANY + || qtype == RR_TYPE_TXT + || qtype == RR_TYPE_NULL) + && qlen > dlen + 1 + HDRSIZE) { - connection *c = conns [client - 1]; - - redo: + // correct class, domain: parse + int client, seqno; + decode_header (qname, client, seqno); - for (vector::iterator i = c->dns_rcvpq.begin (); - i != c->dns_rcvpq.end (); - ++i) - if ((*i)->seqno == seqno) - { - // already seen that request: simply reply with the cached reply - dns_rcv *r = *i; - - printf ("DUPLICATE %d\n", htons (r->pkt->id));//D - - offs = r->pkt->len; - memcpy (pkt->at (0), r->pkt->at (0), offs); - goto duplicate_request; - } + u8 data[MAXSIZE]; + int datalen = cdc62.decode (data, qname + HDRSIZE, qlen - (dlen + 1 + HDRSIZE)); - // new packet, queue - dns_rcv *rcv = new dns_rcv (seqno, data + 3, datalen - 3); - c->dnsv4_receive_rep (rcv); + if (0 < client && client <= conns.size ()) + { + connection *c = conns [client - 1]; + dns_connection *dns = c->dns; + dns_rcv *rcv; + + if (dns) + { + for (vector::iterator i = dns->rcvpq.end (); i-- != dns->rcvpq.begin (); ) + if (SEQNO_EQ ((*i)->seqno, seqno)) + { + // already seen that request: simply reply with the cached reply + dns_rcv *r = *i; + + slog (L_DEBUG, "DNS: duplicate packet received ID %d, SEQ %d", htons (r->pkt->id), seqno); + + // refresh header & id, as the retry count could have changed + memcpy (r->pkt->at (6 * 2 + 1), pkt.at (6 * 2 + 1), HDRSIZE); + r->pkt->id = pkt.id; + + memcpy (pkt.at (0), r->pkt->at (0), offs = r->pkt->len); + + goto duplicate_request; + } + + // new packet, queue + rcv = new dns_rcv (seqno, data, datalen); + dns->receive_rep (rcv); + } - // now generate reply - pkt->ancount = htons (1); // one answer RR - pkt->flags = htons (DEFAULT_SERVER_FLAGS | FLAG_RCODE_OK); + { + pkt [offs++] = 0xc0; pkt [offs++] = 6 * 2; // refer to name in query section - (*pkt) [offs++] = 0xc0; - (*pkt) [offs++] = 6 * 2; // same as in query section + int rtype = dns ? dns->cfg.rrtype : RR_TYPE_A; + pkt [offs++] = rtype >> 8; pkt [offs++] = rtype; // type + pkt [offs++] = RR_CLASS_IN >> 8; pkt [offs++] = RR_CLASS_IN; // class + pkt [offs++] = 0; pkt [offs++] = 0; + pkt [offs++] = 0; pkt [offs++] = dns ? dns->cfg.def_ttl : 0; // TTL - (*pkt) [offs++] = RR_TYPE_TXT >> 8; (*pkt) [offs++] = RR_TYPE_TXT; - (*pkt) [offs++] = RR_CLASS_IN >> 8; (*pkt) [offs++] = RR_CLASS_IN; + int rdlen_offs = offs += 2; - (*pkt) [offs++] = 0; (*pkt) [offs++] = 0; - (*pkt) [offs++] = 0; (*pkt) [offs++] = 0; // TTL + if (dns) + { + int dlen = ntohs (dns->cfg.max_size) - offs; - int dlen = MAX_PKT_SIZE - offs - 2; + // bind doesn't compress well, so reduce further by one label length + dlen -= qlen; - // bind doesn't compress well, so reduce further by one label length - dlen -= qlen; + // only put data into in-order sequence packets, if + // we receive out-of-order packets we generate empty + // replies + //printf ("%d - %d & %x (=%d) < %d\n", seqno, dns->repseq, SEQNO_MASK, (seqno - dns->repseq) & SEQNO_MASK, MAX_WINDOW);//D + if (((seqno - dns->repseq) & SEQNO_MASK) <= MAX_WINDOW) + { + dns->repseq = seqno; + + while (dlen > 1 && !dns->snddq.empty ()) + { + int txtlen = dlen <= 255 ? dlen - 1 : 255; + + if (txtlen > dns->snddq.size ()) + txtlen = dns->snddq.size (); + + pkt[offs++] = txtlen; + memcpy (pkt.at (offs), dns->snddq.begin (), txtlen); + offs += txtlen; + dns->snddq.remove (txtlen); + + dlen -= txtlen + 1; + } + } + + // avoid completely empty TXT rdata + if (offs == rdlen_offs) + pkt[offs++] = 0; - int rdlen_offs = offs += 2; + slog (L_NOISE, "DNS: snddq %d", dns->snddq.size ()); + } + else + { + // send RST + pkt [offs++] = CMD_IP_1; pkt [offs++] = CMD_IP_2; pkt [offs++] = CMD_IP_3; + pkt [offs++] = CMD_IP_RST; + } - while (c->dns_snddq - && !c->dns_snddq->empty () - && dlen > 1) - { - int txtlen = dlen <= 255 ? dlen - 1 : 255; + int rdlen = offs - rdlen_offs; - if (txtlen > c->dns_snddq->size ()) - txtlen = c->dns_snddq->size (); + pkt [rdlen_offs - 2] = rdlen >> 8; + pkt [rdlen_offs - 1] = rdlen; - (*pkt)[offs++] = txtlen; - memcpy (pkt->at (offs), c->dns_snddq->begin (), txtlen); - offs += txtlen; - c->dns_snddq->remove (txtlen); + if (dns) + { + // now update dns_rcv copy + rcv->pkt->len = offs; + memcpy (rcv->pkt->at (0), pkt.at (0), offs); + } + } - dlen -= txtlen + 1; + duplicate_request: ; } + else + pkt.flags = htons (DEFAULT_SERVER_FLAGS | FLAG_RCODE_FORMERR); + } + else if (qtype == RR_TYPE_A + && qlen > dlen + 1 + cdc26.encode_len (sizeof (dns_cfg))) + { + dns_cfg cfg; + cdc26.decode ((u8 *)&cfg, qname, cdc26.encode_len (sizeof (dns_cfg))); + int client = ntohs (cfg.client); - // avoid empty TXT rdata - if (offs == rdlen_offs) - (*pkt)[offs++] = 0; + pkt [offs++] = 0xc0; pkt [offs++] = 6 * 2; // refer to name in query section - int rdlen = offs - rdlen_offs; + pkt [offs++] = RR_TYPE_A >> 8; pkt [offs++] = RR_TYPE_A; // type + pkt [offs++] = RR_CLASS_IN >> 8; pkt [offs++] = RR_CLASS_IN; // class + pkt [offs++] = 0; pkt [offs++] = 0; + pkt [offs++] = 0; pkt [offs++] = cfg.def_ttl; // TTL + pkt [offs++] = 0; pkt [offs++] = 4; // rdlength - (*pkt) [rdlen_offs - 2] = rdlen >> 8; - (*pkt) [rdlen_offs - 1] = rdlen; + slog (L_INFO, _("DNS: client %d connects"), client); - // now update dns_rcv copy - rcv->pkt->len = offs; - memcpy (rcv->pkt->at (0), pkt->at (0), offs); + pkt [offs++] = CMD_IP_1; pkt [offs++] = CMD_IP_2; pkt [offs++] = CMD_IP_3; + pkt [offs++] = CMD_IP_REJ; + + if (0 < client && client <= conns.size ()) + { + connection *c = conns [client - 1]; - duplicate_request: ; + if (cfg.valid ()) + { + pkt [offs - 1] = CMD_IP_SYN; + + delete c->dns; + c->dns = new dns_connection (c); + c->dns->cfg = cfg; + } + } } - else - pkt->flags = htons (DEFAULT_SERVER_FLAGS | FLAG_RCODE_FORMERR); } - pkt->len = offs; + pkt.len = offs; } - - return pkt; } void -vpn::dnsv4_client (dns_packet *pkt) +vpn::dnsv4_client (dns_packet &pkt) { - u16 flags = ntohs (pkt->flags); + u16 flags = ntohs (pkt.flags); int offs = 6 * 2; // skip header - pkt->qdcount = ntohs (pkt->qdcount); - pkt->ancount = ntohs (pkt->ancount); + pkt.qdcount = ntohs (pkt.qdcount); + pkt.ancount = ntohs (pkt.ancount); // go through our request list and find the corresponding request - for (vector::iterator i = dns_sndpq.begin (); + for (vector::iterator i = dns_sndpq.begin (); i != dns_sndpq.end (); ++i) - if ((*i)->pkt->id == pkt->id) + if ((*i)->pkt->id == pkt.id) { - connection *c = (*i)->conn; + dns_connection *dns = (*i)->dns; + connection *c = dns->c; int seqno = (*i)->seqno; u8 data[MAXSIZE], *datap = data; + //printf ("rcv pkt %x\n", seqno);//D + + if ((*i)->retry) + { + dns->send_interval *= 1.01; + if (dns->send_interval > MAX_SEND_INTERVAL) + dns->send_interval = MAX_SEND_INTERVAL; + } + else + { +#if 0 + dns->send_interval *= 0.999; +#endif + // the latency surely puts an upper bound on + // the minimum send interval + double latency = ev_now () - (*i)->sent; + + if (latency < dns->min_latency) + dns->min_latency = latency; + + if (dns->send_interval > dns->min_latency * conf.dns_overlap_factor) + dns->send_interval = dns->min_latency * conf.dns_overlap_factor; + + if (dns->send_interval < conf.dns_send_interval) + dns->send_interval = conf.dns_send_interval; + } delete *i; dns_sndpq.erase (i); - if (flags & (FLAG_RESPONSE | FLAG_OP_MASK | FLAG_TC)) + if (flags & FLAG_RESPONSE && !(flags & FLAG_OP_MASK)) { char qname[MAXSIZE]; - while (pkt->qdcount-- && offs < MAXSIZE - 4) + while (pkt.qdcount-- && offs < MAXSIZE - 4) { - int qlen = pkt->decode_label ((char *)qname, MAXSIZE - offs, offs); + int qlen = pkt.decode_label ((char *)qname, MAXSIZE - offs, offs); offs += 4; // skip qtype, qclass } - while (pkt->ancount-- && offs < MAXSIZE - 10) + while (pkt.ancount-- && offs < MAXSIZE - 10 && datap) { - pkt->decode_label ((char *)qname, MAXSIZE - offs, offs); + int qlen = pkt.decode_label ((char *)qname, MAXSIZE - offs, offs); - u16 qtype = (*pkt) [offs++] << 8; qtype |= (*pkt) [offs++]; - u16 qclass = (*pkt) [offs++] << 8; qclass |= (*pkt) [offs++]; - u32 ttl = (*pkt) [offs++] << 24; - ttl |= (*pkt) [offs++] << 16; - ttl |= (*pkt) [offs++] << 8; - ttl |= (*pkt) [offs++]; + u16 qtype = pkt [offs++] << 8; qtype |= pkt [offs++]; + u16 qclass = pkt [offs++] << 8; qclass |= pkt [offs++]; + u32 ttl = pkt [offs++] << 24; + ttl |= pkt [offs++] << 16; + ttl |= pkt [offs++] << 8; + ttl |= pkt [offs++]; + u16 rdlen = pkt [offs++] << 8; rdlen |= pkt [offs++]; - u16 rdlen = (*pkt) [offs++] << 8; rdlen |= (*pkt) [offs++]; - - if (rdlen <= MAXSIZE - offs) + if (qtype == RR_TYPE_NULL || qtype == RR_TYPE_TXT) { - // decode bytes, finally - - while (rdlen) + if (rdlen <= MAXSIZE - offs) { - int txtlen = (*pkt) [offs++]; + // decode bytes, finally + + while (rdlen) + { + int txtlen = pkt [offs++]; - assert (txtlen + offs < MAXSIZE - 1); + assert (txtlen + offs < MAXSIZE - 1); - memcpy (datap, pkt->at (offs), txtlen); - datap += txtlen; offs += txtlen; + memcpy (datap, pkt.at (offs), txtlen); + datap += txtlen; offs += txtlen; + + rdlen -= txtlen + 1; + } + } + } + else if (qtype == RR_TYPE_A) + { + u8 ip [4]; + + ip [0] = pkt [offs++]; + ip [1] = pkt [offs++]; + ip [2] = pkt [offs++]; + ip [3] = pkt [offs++]; + + if (ip [0] == CMD_IP_1 + && ip [1] == CMD_IP_2 + && ip [2] == CMD_IP_3) + { + slog (L_TRACE, _("DNS: got tunnel meta command %02x"), ip [3]); - rdlen -= txtlen + 1; + if (ip [3] == CMD_IP_RST) + { + slog (L_DEBUG, _("DNS: got tunnel RST request")); + + delete dns; c->dns = 0; + + return; + } + else if (ip [3] == CMD_IP_SYN) + { + slog (L_DEBUG, _("DNS: got tunnel SYN reply, server likes us.")); + dns->established = true; + } + else if (ip [3] == CMD_IP_REJ) + { + slog (L_DEBUG, _("DNS: got tunnel REJ reply, server does not like us, aborting.")); + abort (); + } + else + slog (L_INFO, _("DNS: got unknown meta command %02x"), ip [3]); } + else + slog (L_INFO, _("DNS: got spurious a record %d.%d.%d.%d"), + ip [0], ip [1], ip [2], ip [3]); + return; } + int client, rseqno; + decode_header (qname, client, rseqno); + + if (client != THISNODE->id) + { + slog (L_INFO, _("DNS: got dns tunnel response with wrong clientid, ignoring")); + datap = 0; + } + else if (rseqno != seqno) + { + slog (L_DEBUG, _("DNS: got dns tunnel response with wrong seqno, badly caching nameserver?")); + datap = 0; + } } } // todo: pkt now used - c->dnsv4_receive_rep (new dns_rcv (seqno, data, datap - data)); + if (datap) + dns->receive_rep (new dns_rcv (seqno, data, datap - data)); break; } - - delete pkt; } void -vpn::dnsv4_ev (io_watcher &w, short revents) +vpn::dnsv4_ev (ev::io &w, int revents) { - if (revents & EVENT_READ) + if (revents & EV_READ) { dns_packet *pkt = new dns_packet; struct sockaddr_in sa; socklen_t sa_len = sizeof (sa); - pkt->len = recvfrom (w.fd, &((*pkt)[0]), MAXSIZE, 0, (sockaddr *)&sa, &sa_len); + pkt->len = recvfrom (w.fd, pkt->at (0), MAXSIZE, 0, (sockaddr *)&sa, &sa_len); if (pkt->len > 0) { - if (pkt->flags & htons (FLAG_TC)) + if (ntohs (pkt->flags) & FLAG_RESPONSE) + dnsv4_client (*pkt); + else { - slog (L_WARN, _("DNS request/response truncated, check protocol settings.")); - //TODO connection reset + dnsv4_server (*pkt); + sendto (w.fd, pkt->at (0), pkt->len, 0, (sockaddr *)&sa, sa_len); } - if (THISNODE->dns_port) - { - pkt = dnsv4_server (pkt); - sendto (w.fd, &((*pkt)[0]), pkt->len, 0, (sockaddr *)&sa, sa_len); - } - else - dnsv4_client (pkt); + delete pkt; } } } bool -connection::send_dnsv4_packet (vpn_packet *pkt, const sockinfo &si, int tos) +vpn::send_dnsv4_packet (vpn_packet *pkt, const sockinfo &si, int tos) { - // never initialized - if (!dns_snddq && !dns_rcvdq) - { - dns_rcvdq = new byte_stream (MAX_BACKLOG * 2); - dns_snddq = new byte_stream (MAX_BACKLOG); + int client = ntohl (si.host); - //dns_rcvseq = dns_sndseq = 0; + assert (0 < client && client <= conns.size ()); - dns_si.set (::conf.dns_forw_host, ::conf.dns_forw_port, PROT_DNSv4); - } - - if (!dns_snddq->put (pkt)) - return false; + connection *c = conns [client - 1]; - // start timer if neccessary - if (!THISNODE->dns_port && !dnsv4_tw.active) - dnsv4_cb (dnsv4_tw); + if (!c->dns) + c->dns = new dns_connection (c); + + if (c->dns->snddq.put (pkt)) + c->dns->tw (); + // always return true even if the buffer overflows return true; } void -connection::dnsv4_cb (time_watcher &w) +connection::dnsv4_reset_connection () +{ + //delete dns; dns = 0; //TODO +} + +#define NEXT(w) do { if (next > (w)) next = w; } while (0) + +void +dns_connection::time_cb (ev::timer &w, int revents) { + // servers have to be polled + if (THISNODE->dns_port) + return; + // check for timeouts and (re)transmit - tstamp next = NOW + 60; - dns_req *send = 0; + tstamp next = ev::now () + poll_interval; + dns_snd *send = 0; - for (vector::iterator i = vpn->dns_sndpq.begin (); + for (vector::iterator i = vpn->dns_sndpq.begin (); i != vpn->dns_sndpq.end (); ++i) { - dns_req *r = *i; + dns_snd *r = *i; - if (r->next <= NOW) + if (r->timeout <= ev_now ()) { if (!send) { send = r; - if (r->retry)//D - printf ("req %d, retry %d\n", r->pkt->id, r->retry); r->retry++; - r->next = NOW + r->retry; + r->timeout = ev_now () + (r->retry * min_latency * conf.dns_timeout_factor); + //printf ("RETRY %x (%d, %f)\n", r->seqno, r->retry, r->timeout - ev_now ());//D + + // the following code changes the query section a bit, forcing + // the forwarder to generate a new request + if (r->stdhdr) + encode_header ((char *)r->pkt->at (6 * 2 + 1), THISNODE->id, r->seqno, r->retry); } } - - if (r->next < next) - next = r->next; + else + NEXT (r->timeout); } - if (!send - && vpn->dns_sndpq.size () < MAX_OUTSTANDING) + if (!send) { - send = new dns_req (this); - send->gen_stream_req (dns_sndseq, dns_snddq); - vpn->dns_sndpq.push_back (send); + // generate a new packet, if wise - dns_sndseq = (dns_sndseq + 1) & SEQNO_MASK; - } + if (!established) + { + if (vpn->dns_sndpq.empty ()) + { + send = new dns_snd (this); - tstamp min_next = NOW + (1. / (tstamp)MAX_RATE); + cfg.reset (THISNODE->id); + send->gen_syn_req (); + } + } + else if (vpn->dns_sndpq.size () < conf.dns_max_outstanding + && !SEQNO_EQ (rcvseq, sndseq - (MAX_WINDOW - 1))) + { + if (last_sent + send_interval <= ev_now ()) + { + //printf ("sending data request etc.\n"); //D + if (!snddq.empty () || last_received + 1. > ev_now ()) + { + poll_interval = send_interval; + NEXT (ev_now () + send_interval); + } - if (send) - { - dns_packet *pkt = send->pkt; + send = new dns_snd (this); + send->gen_stream_req (sndseq, snddq); + send->timeout = ev_now () + min_latency * conf.dns_timeout_factor; + //printf ("SEND %x (%f)\n", send->seqno, send->timeout - ev_now (), min_latency, conf.dns_timeout_factor);//D - next = min_next; + sndseq = (sndseq + 1) & SEQNO_MASK; + } + else + NEXT (last_sent + send_interval); + } + + if (send) + vpn->dns_sndpq.push_back (send); + } - sendto (vpn->dnsv4_fd, &((*pkt)[0]), pkt->len, 0, dns_si.sav4 (), dns_si.salenv4 ()); + if (send) + { + last_sent = ev_now (); + sendto (vpn->dnsv4_fd, + send->pkt->at (0), send->pkt->len, 0, + vpn->dns_forwarder.sav4 (), vpn->dns_forwarder.salenv4 ()); } - else if (next < min_next) - next = min_next; - w.start (next); + slog (L_NOISE, "DNS: pi %f si %f N %f (%d:%d %d)", + poll_interval, send_interval, next - ev_now (), + vpn->dns_sndpq.size (), snddq.size (), + rcvpq.size ()); + + // TODO: no idea when this happens, but when next < ev_now (), we have a problem + // doesn't seem to happen anymore + if (next < ev_now () + 0.001) + next = ev_now () + 0.1; + + w.start (next - ev_now ()); } #endif